On 25/09/2020 12:06, Weiyi Lu wrote:
On Thu, 2020-09-10 at 19:28 +0200, Enric Balletbo i Serra wrote:
Dear all,
This is a new driver with the aim to deprecate the mtk-scpsys driver.
The problem with that driver is that, in order to support more Mediatek
SoCs you need to add some logic to handle properly the power-up
sequence of newer Mediatek SoCs, doesn't handle parent-child power
domains and need to hardcode all the clocks in the driver itself. The
result is that the driver is getting bigger and bigger every time a
new SoC needs to be supported.
Hi Enric and Matthias,
First of all, thank you for the patch. But I'm worried the problem you
mentioned won't be solved even if we work on this new driver in the
future. My work on the MT8183 scpsys(now v17) is to implement the new
hardware logic. Here, I also see related patches, which means that these
new logics are necessary. Why can't we work on the original driver?
Well the decision was to change the driver in a not compatible way to make
device tree entries better. If we work on the old driver, we would need to find
some creative ways to handle old bindings vs new bindings.
So I thought it would be better doing a fresh start implementing mt1873 support
for reference and add mt8183 as new SoC. From what I have seen mt8192 and others
fit the driver structure too.
Meanwhile, I thought maybe we should separate the driver into general
control and platform data for each SoC, otherwise it'll keep getting
bigger and bigger if it need to be support new SoC.
We could in a later series split the SoC depended data structures and put them
in drivers/soc/mediatek/pm-domains-mt8183.h or something like this. Is that what
you mean?
And consider DVFSRC
(dynamic voltage and frequency scaling resource collector), should we
keep the original driver name "scpsys" instead of "pm-domains" because
it may provide more functions than power domains?
It's on my list to look deeper into this series. The thing with the new driver
is, that the binding takes into account, that scpsys has several hardware block,
which are represented as child nodes in DTS. The pm-domains is just one of these
functionalities and I think DVFSRC should be a new driver with a child node of
scpsys in DTS. Does this make sense?
Regards,
Matthias
All this information can be getted from a properly defined binding, so
can be cleaner and smaller, hence, we implemented a new driver. For
now, only MT8173 and MT8183 is supported but should be fairly easy to
add support for new SoCs.
Best regards,
Enric
Enric Balletbo i Serra (4):
dt-bindings: power: Add bindings for the Mediatek SCPSYS power domains
controller
soc: mediatek: Add MediaTek SCPSYS power domains
arm64: dts: mediatek: Add mt8173 power domain controller
dt-bindings: power: Add MT8183 power domains
Matthias Brugger (8):
soc: mediatek: pm-domains: Add bus protection protocol
soc: mediatek: pm_domains: Make bus protection generic
soc: mediatek: pm-domains: Add SMI block as bus protection block
soc: mediatek: pm-domains: Add extra sram control
soc: mediatek: pm-domains: Add subsystem clocks
soc: mediatek: pm-domains: Allow bus protection to ignore clear ack
soc: mediatek: pm-domains: Add support for mt8183
arm64: dts: mediatek: Add mt8183 power domains controller
.../power/mediatek,power-controller.yaml | 173 ++++
arch/arm64/boot/dts/mediatek/mt8173.dtsi | 78 +-
arch/arm64/boot/dts/mediatek/mt8183.dtsi | 160 +++
drivers/soc/mediatek/Kconfig | 13 +
drivers/soc/mediatek/Makefile | 1 +
drivers/soc/mediatek/mtk-infracfg.c | 5 -
drivers/soc/mediatek/mtk-pm-domains.c | 952 ++++++++++++++++++
include/dt-bindings/power/mt8183-power.h | 26 +
include/linux/soc/mediatek/infracfg.h | 39 +
9 files changed, 1433 insertions(+), 14 deletions(-)
create mode 100644 Documentation/devicetree/bindings/power/mediatek,power-controller.yaml
create mode 100644 drivers/soc/mediatek/mtk-pm-domains.c
create mode 100644 include/dt-bindings/power/mt8183-power.h
